Large Language Model-Based Autonomous Agents: Trends and Directions

Abstract

This paper explores the vibrant field of autonomous agents based on large language models. In recent years transformer-based large language models (LLMs) have advanced state of the art considerably in a wide range of natural language tasks and demonstrated almost human-like reasoning capabilities and world knowledge. Since autonomous agents rely on such properties, advances in LLMs have accelerated the progress in autonomous agents. This paper reviews the literature by briefly describing how LLMs work and how they can be leveraged in the overall architecture of an autonomous agent to produce significantly more capable and robust agents. Planning, memory, and action components of the autonomous agent are examined separately and a discussion of trends and future directions follows.